config.yml

data:
  # The name of the dataset, either mnist or frey
  dataset_name: frey
  # The number of sample images to show before training
  n_imgs: 4
vae:
  # The number of nodes in the hidden layer of the encoder network
  encoder_hidden_dim: 200
  # The number of dimensions of the latent vector
  latent_dim: 20
  # The number of nodes in the hidden layer of the decoder network
  decoder_hidden_dim: 200
  # The name of the VAE class, either VAE_EGDG (encoder Gaussian, decoder Gaussian, for frey) or
  # VAE_EGDB (encoder Gaussian, decoder Bernoulli, for mnist)
  name: VAE_EGDG
train:
  # The learning rate of the optimizer.
  lr: 0.001
  # The number of epochs to train for.
  epochs: 10000
  # The number of instances in a mini-batch of the optimizer.
  batch_size: 100
plot:
  # Whether to invert the ELBO values (we minimize what they maximize in the paper, so to get similar plots, we
  # need to invert them.
  neg_elbo_values: True
  # The x-axis label and type, either epochs or samples
  x_axis_label: samples
  # The scale of the x-axis, either linear or log
  x_axis_scale: log
  # The min of the y-lim in plotting the ELBO
  ylim_min: 0
  # The max of the y-lim in plotting the ELBO
  ylim_max: 1800
  # The number of reconstructed images to plot. 10 is the most space-wise.
  n_imgs_rec: 10